Mistral Small 3 24b is the first model under 70b I’ve seen pass the “apple” test (even using Q4).

> You modified the standard apple test prompt by adding the “coherent”part. yeah thats not how it works, put shit in get shit out, work on prompting better. >Secondly, you modified the default model temperature to 1.3 which is way past default. There is no default. It's per model. Some work for some models some don't. For example there's models that if you put repeat_panalty they become incoherent. >I personally just don’t like trying to make a model do something by changing parameters to pass a test Irrelevant what you like, they are mathematical equations, you need to tune them to the task you want executed. Imagine wanting to write a book and habing a low temperature, that is idiotic